[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Working On Noise
- To: firstname.lastname@example.org, email@example.com
- Subject: Working On Noise
- From: Tom Droege <firstname.lastname@example.org>
- Date: Tue, 13 Oct 1998 15:12:23 -0500
- Date-warning: Date header was inserted by FNAL.FNAL.GOV
- Old-Return-Path: <email@example.com>
- Resent-Date: Tue, 13 Oct 1998 16:15:42 -0400
- Resent-From: firstname.lastname@example.org
- Resent-Message-ID: <"NLb5ZC.A.JjD.tQ7I2"@kani.wwa.com>
- Resent-Sender: email@example.com
I hung Norman's poem up on my wall over the workbench, and the noise dropped
to near zero. Later I discovered it was still there, but at least I now
have a better understanding of what is going on. There is an oscillation to
be found, but not quite the usual sort. I will try to explain.
I have been trying to take dark exposures of various lengths. The idea is
that by extrapolating the mean values back to an exposure of zero length,
one can arrive at the dark current. One component of the noise is just due
to the number of electrons. So one component of the noise should increase
as the square root of the exposure. At long exposures, this term should be
dominant, so one should be able to tie the number of electrons to the noise
for a long exposure, then work back and get the electronic noise by the
extrapolation to zero exposure. For comparison, I have an electronic noise
measurement made by running the vertical shift backwards. This should give
a measurement of everything but the signal due to the dark current electrons.
Attempts to take data for this measurement have failed. At short exposures,
everything looks great. I can take a one minute and a two minute dark frame
and find the zero current base line and get values for electronic and dark
current noise that make sense. Any longer exposures show something really
funny. Instead of getting a nice distribution, one sees three distributions
with a clean break between them. i.e. no counts. This is similar to the
case of a "normal" oscillation. Since a sine wave oscilation on top of a
signal spends more time at the peaks that between, one gets a tri - peak
distribution for an oscillating system. But the peaks are not separate,
there are some points between peaks. The distributions measured have clean
breaks between peaks.
A Theory of What is Going On.
Suppose bit n of the ADC output is feeding back into the input. If n is a
fairly large bit, it may not change when the exposure is short - i.e. the
noise is low. Once the noise gets higher, this bit is changing al the time
and now feeds back into the input. Now there are preffered codes in the
output, and we get separated peaks.
OK, folks, stay tuned to see what is really going on. This is the kind of
thing that I track down for a living. Arn't you glad you write code and
create "objects" which if they have feedback, do so with digital decision,
and can be tracked down. Here I have to deal with electric fields and
currents, and the parts of the circuit that never appear on the drawing.
At 12:29 PM 10/12/98 -0700, you wrote:
>> Boy, the list is quiet. Here I am in the pit, fighting three Ox Men
>> from Gore with my bear hands and brains, and I hear nary a cheer.
>> It is OK to shout encouragement.
>> So far, I am not so damaged and might yet will this battle.
>Three cheers for Thee, O Tom,
>heroic slayer of the Ox Men from Gore!
>Thee are battling Noise, Hum & Heat,
>fearsome threesome who would destroy
>our hoped-for images of sky & stars.
>Here we stand acclaiming Thee
>O wizard and knight of the hot iron,
>who fights to keep our nights
>as dark as an electron.
>Three cheers for Tom,
>you TASSers and bystanders,
>three cheers for Tom,
>as in the pit for us he stands, er,
>fighting Noise, Hum and Heat.
>(s) Norman "rye ming in prose" Molhant